A neural network approach to detect traffic anomalies in a communication network

LEAPR (Line Element Anomaly Pattern Recognition) is a Knowledge Source (KS) used in a multiparadigm diagnostic system. This KS was trained off-line with traffic data from a simulated Wide Area Network (WAN) under normal and abnormal network operations. Abnormal network operations were characterized by different traffic increases and normal network operations were characterized by no traffic increase. After training, this KS was able to rapidly classify in real time the monitored traffic pattern as either NORMAL or ABNORMAL. After diagnosing an ABNORMAL traffic pattern, the LEAPR KS warns the other KSs used in the multiparadigm diagnostic system of an abnormal traffic increase. An abnormal traffic increase might be caused by a congestion or a hardware fault in the network that causes the traffic to be re-routed to a specific line.;This thesis presents different LEAPR KS prototypes with their respective results. The different prototypes led to the implementation of the LEAPR KS that diagnoses ABNORMAL traffic patterns of any router of the simulated WAN used in the Pegasus research project.